Research On Collaborative Filtering Recommendation Algorithm Based On Hybrid User Model

Widely using of Internet and rapidly development of E-commerce caused information overload, which made difficulties for consumers to find their needed products within a mass of product information, thus E-commerce recommender systems emerge as the times require. Today, E-commerce recommender systems are immature in practical use, and still have a lot of problems, like the quality of recommendation being seriously depressed by enormous and sparse ratings of consumers, bad system expansibility, bad recommendation real-time, etc. To solve these main problems of current recommender systems, this dissertation valuably explores and researches the key techniques of user model and collaborative filtering algorithms in E-commerce personalized recommender systems.Collaborative filtering is the most widely used and successful technology for personalized recommender systems. However it faces challenges of scalability and recommendation accuracy. Collaborative filtering can be divided into memory based and model based. The former is more accurate while the latter performs better in scalability. This paper proposes a hybrid user model. The recommender system based on this model not only holds the advantage of recommendation accuracy in memory-based method, but also has the scalability as good as model-based method.In the aspect of user model, the dissertation analyses defects of classical user model of collaborative filtering recommendation. And hybrid user model is constructed based on item content descriptions and demographic information. The hybrid user model condenses item content description, demographic information and user-item rating matrix, which raises the density of data and helps to solve the problems of data sparsity and hard rating obtainment. Feature interest measure is introduced in the hybrid user model, which can reflect the degree of feature preference of users and obtain more accurate similarity between target user and the neighbors.In the aspect of collaborative filtering, this dissertation analyses sparsity, scalability, real-time and recommendation accuracy issues of collaborative filtering algorithms in current E-commerce personalized recommender systems. To solve these problems, collaborative filtering recommendation algorithm based on hybrid user model is proposed. The algorithm adopts a combination filtering method which firstly constructs a user model offline by combining filtering technologies based on content and demographic information, then makes recommendation online on the basis of the model by using collaborative filtering. Combination is introduced at three different layers: feature layer, model layer and collaborative filtering algorithm layer, which reduces system complexity, shortens the computing time and improves scalability and recommendation accuracy. Genetic algorithm is introduced in collaborative filtering algorithm layer, which learns the weight features in hybrid user model and significantly enhances the recommendation accuracy due to the accurate description of user preference.We do imitation experiments on improved algorithm proposed by the dissertation by means of MovieLens data set. Experimental results show that collaborative filtering recommendation algorithm based on hybrid user model are better than experiment contrasted algorithm in the aspect of accuracy, integrality, scalability in recommendation.